AI Operations
AI Cost Ratio
AI Cost Ratio is the percentage of a company's revenue spent on AI compute — LLM API calls, inference costs, model training, and AI infrastructure. It's the new version of "cost of goods sold" for AI-run companies.
Formula
AI Cost Ratio = (Total AI compute costs / Revenue) × 100Benchmarks
| AI Cost Ratio | Assessment |
|---|---|
| < 5% | Excellent — highly efficient AI usage |
| 5–10% | Good — standard for AI-run SaaS |
| 10–20% | Acceptable — may need optimization |
| 20–30% | Concerning — AI costs eating margins |
| 30%+ | Unsustainable — rethink architecture |
Typical AI Costs (2026 Pricing)
| Operation | Cost | Example |
|---|---|---|
| Claude Sonnet API call | ~$0.003–0.015 per call | Customer support response |
| GPT-4 API call | ~$0.005–0.03 per call | Content generation |
| Image generation | ~$0.02–0.05 per image | Marketing visuals |
| Code generation | ~$0.01–0.05 per task | Bug fix, feature |
| Embedding / search | ~$0.0001 per query | Knowledge retrieval |
AI Cost Optimization
- Model selection — Use smaller models for simple tasks, large models for complex ones
- Caching — Cache frequent queries to avoid redundant API calls
- Batching — Process multiple requests together for volume discounts
- Prompt optimization — Shorter, more efficient prompts reduce token costs
- Self-hosted models — For high-volume tasks, running open-source models can be cheaper
AI Cost Ratio vs Traditional SaaS Costs
| Cost Category | Traditional SaaS | AI-Run SaaS |
|---|---|---|
| Engineering | 25–35% of revenue | 2–5% (AI compute) |
| Marketing | 20–30% of revenue | 1–3% (AI compute) |
| Support | 10–15% of revenue | 0.5–2% (AI compute) |
| Infrastructure | 5–10% of revenue | 5–10% (same) |
| Total | 60–90% | 8–20% |
The AI cost ratio replaces three separate human cost categories with one much smaller compute cost.
On EvolC, we track AI cost ratio as a key efficiency metric — it shows how well a company uses AI and how much margin flows to investors.